Information processing device, information processing system, information processing method, and program

Abstract

To increase the accuracy of evaluating a global model.SOLUTION: An information processing apparatus of an embodiment has a first acquisition unit, a second acquisition unit, a weight calculation unit, and an index value calculation unit. The first acquisition unit acquires a global model based on a dataset related to a plurality of cohorts. The second acquisition unit acquires a local model based on a first dataset related to at least one first cohort of the plurality of cohorts. The weight calculation unit calculates a weight based on a first data distribution related to the first dataset and a second data distribution related to a second dataset related to the plurality of cohorts. The index value calculation unit calculates an index value related to the global model on the basis of the first dataset and the weight.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The embodiments disclosed in this specification and the drawings relate to an information processing device, an information processing system, an information processing method, and a program. [Background technology]

[0002] In the medical field, data held by clients (also called sites) is highly confidential, so distributed learning (online learning), which allows models to be built without each client directly sharing their data, is a learning method highly compatible with the medical field. Distributed learning is conducted using data from multiple clients. For example, distributed learning uses a model trained by each client based on a dataset related to a cohort such as a patient (local model), and a model trained using a variety of data by integrating each client's model (global model).

[0003] A global model is a model based on datasets related to multiple cohorts. For example, the global model is generated by a central server based on local models provided by multiple clients and provided to each client. The purpose of building a global model is to create a model trained on various data so that all participating clients can use a high-performance model.

[0004] However, in conventional distributed learning, it is not possible to evaluate a global model by taking into account trends in all data from clients participating in the distributed learning, and model evaluation is performed, for example, on a client-by-client basis, making it difficult to fully evaluate the performance of the global model used by each client. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Special Publication No. 2020-525892 [Patent Document 2] Special Publication No. 2019-526851 [Patent Document 3] Special Publication No. 2020-530159 Summary of the Invention [Problem to be solved by the invention]

[0006] The problem to be solved by the embodiments disclosed in this specification and the drawings is to improve the accuracy of evaluating a global model. However, the problem to be solved by the embodiments disclosed in this specification and the drawings is not limited to the above problem. Problems corresponding to the effects of each configuration shown in the embodiments described below can also be positioned as other problems. [Means for solving the problem]

[0007] An information processing apparatus according to an embodiment includes a first acquisition unit, a second acquisition unit, a weight calculation unit, and an index value calculation unit. The first acquisition unit acquires a global model based on datasets related to multiple cohorts. The second acquisition unit acquires a local model based on a first dataset related to at least one first cohort among the multiple cohorts. The weight calculation unit calculates weights based on a first data distribution related to the first dataset and a second data distribution related to a second dataset related to the multiple cohorts. The index value calculation unit calculates an index value related to the global model based on the first dataset and the weights. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a configuration diagram of an information processing system 1 according to an embodiment. [Figure 2] Site 100 configuration diagram. [Figure 3] FIG. 2 is a diagram illustrating the configuration of a central server 200. [Figure 4] FIG. 2 is a sequence diagram showing the processing flow of the site 100 and the central server 200. [Figure 5] FIG. 10 is a diagram showing an image of calculation and transmission of data distribution. [Figure 6] A diagram showing an image of how to calculate the global density ratio. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, an information processing apparatus, an information processing system, an information processing method, and a program according to embodiments will be described with reference to the drawings.

[0010] 1 is a configuration diagram of an information processing system 1 according to an embodiment. The information processing system 1 includes, for example, a plurality of sites 100 and a central server 200. The plurality of sites 100 are connected to the central server 200 via a communication network NW so as to be able to transmit and receive data. The central server 200 generates a global model through distributed learning using learning data acquired by the plurality of sites 100 and provides the global model to the plurality of sites 100.

[0011] The communication network NW refers to all information and communication networks that utilize telecommunications technology. It includes wireless / wired LANs such as hospital backbone LANs (Local Area Networks), the Internet, telephone communication networks, optical fiber communication networks, cable communication networks, and satellite communication networks.

[0012] The site 100 represents an information processing device installed in, for example, a medical facility. The site 100, for example, collects and provides information related to treatment and diagnosis at the medical facility. The central server 200 is installed in, for example, a facility other than the medical facility. The central server 200 collects information for the site 100, analyzes the collected information, generates information to be provided to the medical facility, and transmits the generated information to the site 100 to provide it to the medical facility. The central server 200 may be installed in the medical facility.

[0013] 2 is a configuration diagram of the site 100. The site 100 includes, for example, a communication interface 110, an input interface 120, a processing circuit 130, and a memory 140. The communication interface 110 communicates with an external device, for example, a central server 200, via a communication network NW. The communication interface 110 includes, for example, a communication interface such as a network interface card (NIC).

[0014] The input interface 120 accepts various input operations by an operator and outputs an electrical signal indicating the content of the accepted input operation to the processing circuit 130. The input interface 120 is realized by, for example, a mouse, a keyboard, a touch panel, a drag ball, a switch, a button, a joystick, a camera, an infrared sensor, a microphone, etc.

[0015] In this specification, the input interface is not limited to an interface having physical operation parts such as a mouse, keyboard, etc. For example, an example of an input interface also includes an electrical signal processing circuit that receives an electrical signal corresponding to an input operation from an external input device provided separately from the device and outputs this electrical signal to a control circuit.

[0016] The processing circuit 130 has a processor such as a CPU (Central Processing Unit), for example. The processing circuit 130 controls the overall operation of the site 100. The processing circuit 130 includes, for example, a first acquisition function 131, a second acquisition function 132, a learning function 133, a local data distribution calculation function 134, an index value calculation function 136, a first selection function 137, and a second selection function 138. The processing circuit 130 realizes these functions by, for example, a hardware processor executing a program stored in a storage device (storage circuit).

[0017] The term "hardware processor" refers to a circuit such as a central processing unit (CPU), a graphics processing unit (GPU), an application specific integrated circuit (ASIC), a programmable logic device (e.g., a simple programmable logic device (SPLD) or a complex programmable logic device (CPLD)), or a field programmable gate array (FPGA). Instead of storing a program in a memory device, the program may be directly embedded in the circuit of the hardware processor. In this case, the hardware processor realizes its function by reading and executing the program embedded in the circuit. The hardware processor is not limited to being configured as a single circuit, but may be configured as a single hardware processor by combining multiple independent circuits to realize each function. The memory device may be a non-transitory (hardware) storage medium. Furthermore, multiple components may be integrated into a single hardware processor to realize each function.

[0018] Each component of the processing circuitry 130 may be distributed and realized by multiple pieces of hardware. The processing circuitry 130 may not be a component of the site 100, but may be realized by a processing device capable of communicating with the site 100. Each function included in the processing circuitry 130 may be distributed among multiple circuits, or may be made available by launching application software stored in the memory 140.

[0019] The memory 140 is included in a storage device. The memory 140 is provided with a training data storage area 141, a global model storage area 142, a local model storage area 143, a trained global model storage area 144, and a trained local model storage area 145. The training data storage area 141 stores training data for generating a local model. The training data includes a dataset related to a cohort of patients or the like in a medical facility where the site 100 is located. This training data is, for example, data used when the most recent local model was generated. The cohort of patients or the like in the medical facility where the site 100 is located is an example of a first cohort. The dataset related to the first cohort is an example of a first dataset.

[0020] The global model storage area 142 stores multiple global models. The local model storage area 143 stores multiple local models. The trained global model storage area 144 stores multiple trained global models. A trained global model is a model obtained by training a global model using the training data stored in the training data storage area 141. The trained local model storage area 145 stores multiple trained local models. A trained local model is a model obtained by training a local model using the training data stored in the training data storage area 141.

[0021] The first acquisition function 131 acquires a global model that is sent and provided by the central server 200 and received by the communication interface 110. The first acquisition function 131 stores the acquired global model in a global model storage area 142 of the memory 140. As the first acquisition function 131 stores a global model, the number of global models stored in the global model storage area 142 increases each time.

[0022] The second acquisition function 132 acquires the model selected by the second selection function 138 as a local model. The second acquisition function 132 stores the acquired local model in a local model storage area 143 of the memory 140. As the second acquisition function 132 stores the local model, the number of local models stored in the local model storage area 143 increases each time.

[0023] The learning function 133 uses the learning data stored in the learning data storage area 141 of the memory 140 to learn the global model acquired by the first acquisition function 131 by machine learning, deep learning, or the like, to generate a trained global model. The learning function 133 stores the generated trained global model in the trained global model storage area 144 of the memory 140. As the learning function 133 stores the trained global model, the number of trained global models stored in the trained global model storage area 144 increases each time. The learning function 133 is an example of a learning unit.

[0024] The learning function 133 uses the learning data stored in the learning data storage area 141 of the memory 140 to learn the local model acquired by the second acquisition function 132 by machine learning, deep learning, or the like, to generate a trained local model. The learning function 133 stores the generated trained local model in the trained local model storage area 145 of the memory 140. As the learning function 133 stores the trained local model, the number of trained local models stored in the trained local model storage area 145 increases each time.

[0025] The local data distribution calculation function 134 calculates the local data distribution of the dataset related to the cohort based on the training data stored in the training data storage area 141 of the memory 140. The local data distribution calculation function 134 transmits the calculated local data distribution to the central server 200 using the communication interface 110. The local data distribution is an example of a first data distribution.

[0026] The weight calculation function 135 calculates weights to be used when calculating index values ​​related to the global model based on the local data distribution calculated by the local data distribution calculation function 134 and the global data distribution transmitted by the central server 200. The weights will be explained later.

[0027] The index value calculation function 136 calculates index values ​​for the global model, local model, trained global model, and trained local model (hereinafter referred to as model group) stored in the memory 140. The index value for the global model is, for example, an index value that serves as a criterion for selecting a model (hereinafter referred to as provided model) to be provided to the central server 200 as a model used to generate the global model, or indicates the suitability of the model as a model to be used in medical practice at the site 100 (hereinafter referred to as operational model). The index value calculation function 136 is an example of an index value calculation unit. The operational model is an example of an operational local model.

[0028] The first selection function 137 selects a model to be provided from a group of models including a global model and a trained global model, based on the index value of each model in the group of models calculated by the index value calculation function 136. The first selection function 137 transmits the selected model to be provided to the central server 200 using the communication interface 110. The first selection function 137 is an example of a first selection unit.

[0029] The second selection function 138 selects an operation model from a group of models stored in the memory 140. The site 100 uses the operation model selected by the second selection function 138 to operate various processes executed within the site 100. The operation model may be selected using a known model evaluation method as appropriate. The second selection function 138 is an example of a second selection unit.

[0030] 3 is a configuration diagram of the central server 200. The central server 200 is provided, for example, in a management facility that centrally manages information provided by and to multiple sites 100. The central server 200 includes, for example, a communication interface 210, an input interface 220, a processing circuit 230, and a memory 240. The communication interface 210 communicates with external devices, for example, the multiple sites 100, via a communication network NW.

[0031] The communication interface 210 includes a communication interface such as a NIC. Distribution data and provided models are transmitted from multiple sites 100 to the central server 200. Therefore, multiple pieces of distribution data and provided models are transmitted to the central server 200. The input interface 220 receives various input operations by an operator and outputs an electrical signal indicating the content of the received input operation to the processing circuit 230.

[0032] The processing circuitry 230 has a processor such as a CPU, for example. The processing circuitry 230 controls the overall operation of the central server 200. The processing circuitry 230 has, for example, a global data distribution calculation function 231 and an integration function 232. The processing circuitry 230 realizes these functions by, for example, a hardware processor executing a program stored in a storage device (storage circuit).

[0033] Each component of the processing circuitry 230 may be distributed and realized by multiple pieces of hardware. The processing circuitry 230 may not be a component of the central server 200, but may be realized by a processing device capable of communicating with the central server 200. Each function included in the processing circuitry 230 may be distributed to multiple circuits, or may be made available by launching application software stored in the memory of the central server 200. Information such as global data distribution and global models is stored in the memory 240.

[0034] The global data distribution calculation function 231 calculates a global data distribution based on the local data distributions of the multiple data sets for each of the multiple cohorts transmitted by the multiple sites 100. The global data distribution is a distribution that aggregates the local data distributions of the multiple data sets. The multiple data sets for the multiple cohorts are an example of a second data set. The global data distribution is an example of a second data distribution.

[0035] The global data distribution calculation function 231 transmits the calculated global data distribution to each of the multiple sites 100 using the communication interface 210. The site 100 that receives the global data distribution calculates a weight using the received global data distribution in the weight calculation function 135. When transmitting the global data distribution to the multiple sites 100, the global data distribution calculation function 231 transmits the global model stored in the memory 240 together with the global data distribution.

[0036] The integration function 232 integrates multiple provided models transmitted by multiple sites 100 to generate a new global model. When storing the newly generated global model in the memory 240, the integration function 232 erases the global models that have already been stored. The integration function may store the newly generated global model without erasing the global models that have already been stored. The integration function 232 is an example of an integration unit.

[0037] Next, the processing in the information processing system 1 will be described. Figure 4 is a sequence diagram showing the processing flow of the site 100 and the central server 200. For example, in the central server 200, distributed learning is started when an operator operates the input interface 220 to indicate the start of distributed learning (step S101). When distributed learning starts, the central server 200 uses the communication interface 210 to send start information to each of the multiple sites (step S103).

[0038] Upon receiving the start information, the site 100 acquires the learning data stored in the learning data storage area 141 of the memory 140 in the local data distribution calculation function 134 (step S201). Subsequently, the site 100 calculates the local data distribution of various data included in the learning data in the local data distribution calculation function 134 (step S203).

[0039] The local data distribution is, for example, a distribution obtained when data included in the learning data is aggregated by item. Any item may be used, such as patient attributes such as age, sex, and physique, the doctor in charge, disease name, symptoms, date and time, and season. The local data distribution calculation function 134 transmits the calculated local data distribution to the central server 200 using the communication interface 110 (step S205).

[0040] The local data distribution calculation function 134 calculates the local data distribution by performing, for example, parametric estimation or non-parametric estimation on the training data. A plurality of local data distributions are transmitted from a plurality of sites 100 to the central server 200. The central server 200, in the global data distribution calculation function 231, compiles the transmitted plurality of local data distributions and calculates the global data distribution (step S105).

[0041] The global data distribution calculation function 231 uses the communication interface 210 to transmit the calculated global data distribution to each site 100 (step S107). When transmitting the global data distribution, the global data distribution calculation function 231 also transmits the global model stored in the memory 240 to each site 100.

[0042] 5 is a diagram illustrating an image of calculation and transmission of data distribution. Each site 100 calculates a local data distribution based on the training data stored in its memory 140. For example, in the first medical facility 300A, a first local data distribution 180A is calculated by parametric estimation or non-parametric estimation based on the first training data 150A stored in the first memory 140A.

[0043] Similarly, at the second medical facility 300B, a second local data distribution 180B is calculated by parametric estimation or non-parametric estimation based on the second training data 150B stored in the second memory 140B. Both the first local data distribution 180A and the second local data distribution 180B are represented by, for example, a probability density function.

[0044] The first local data distribution 180A and the second local data distribution 180B are both transmitted to the central server 200. The global data distribution calculation function 231 of the central server 200 calculates a global data distribution 280 based on the transmitted first local data distribution 180A and second local data distribution 180B. The global data distribution 280 calculated by the central server 200 is transmitted to the sites 100 located in the first medical facility 300A and the second medical facility 300B.

[0045] Returning to Fig. 4, the site 100 receives the global data distribution and the global model transmitted by the central server 200. The site 100, in the learning function 133, updates the received global model by learning such as machine learning using the learning data stored in the learning data storage area 141 of the memory 140 (step S207) to generate a learned global model.

[0046] At this time, the learning function 133 updates the most recent local model among the local models stored in the local model storage area 143 by learning, such as machine learning, using the learning data stored in the learning data storage area 141 (step S207), to generate a trained local model. The trained local model may be generated at a different time from the generation of the trained global model. The learning function 133 stores the generated trained global model and trained local model in the trained global model storage area 144 and the trained local model storage area 145, respectively.

[0047] Next, the site 100 calculates a global density ratio as a weight in the weight calculation function 135 based on the global data distribution transmitted by the central server 200 and the local data distribution stored in the memory 140 (step S209). The weight calculation function 135 calculates the global density ratio by taking the ratio between the global data distribution and the local data distribution.

[0048] 6 is a diagram showing an image of calculating a global density ratio. The central server 200 transmits a global data distribution 260 calculated by the global data distribution calculation function 231 to a site 100 provided in a medical facility 300. The site 100 stores a local data distribution 160. The weight calculation function 135 calculates a global density ratio 360 by taking the ratio of the two distributions, with the global data distribution as the denominator and the local data distribution as the numerator.

[0049] Next, the index value calculation function 136 calculates the error of each model using the training data stored in the training data storage area 141 and the weights calculated by the weight calculation function 135, for example, by importance-weighted cross-validation. The global density ratio corresponds to the importance in importance-weighted cross-validation. By using importance-weighted cross-validation, the local data distributions 160 stored in the memory 140 of the site 100 can be brought closer to the global data distribution 260, while the model error can be calculated.

[0050] The importance-weighted cross-validation method will now be described. The index value calculation function 136 first randomly divides the training data T constituting the local data distribution into m subsets of a size with no or almost no overlap. The training data T is expressed by the following formula (1), and the subsets are expressed by the following formula (2).

[0051]

number

[0052] In the above formula (1), x represents input data and y represents output data.

[0053] After dividing the learning data T, the index value calculation function 136 repeats the following processes (a) and (b) for i=1, 2, . . . , m. (a) Learning data T shown in the following equation (3) i Using all data except for i Ask for. (b) The saved training data T i The generalized error Gi (hat) is calculated using the following equation (4).

[0054]

number

[0055] In the above equation (4), |T i | is a set T i represents the number of elements, and ω(x) represents the importance (global density ratio).

[0056] The index value calculation function 136 calculates the divided learning data T i After calculating the generalized error for each, all the training data T i An index value is calculated by using the average value of the generalized errors as the final predicted value of the generalized error (step S211). The average value of the generalized error is expressed by the following formula (5). The average value of the generalized error (predicted value of the generalized error) becomes the index value when selecting a model to be provided.

[0057]

number

[0058] The index value calculation function 136 calculates an index value for each model included in the model group. The first selection function 137 compares the average values ​​of the generalized errors calculated for each model. The first selection function 137 selects the model with the smallest average value of the generalized errors as the model to be provided (step S213).

[0059] After selecting a provision model, the first selection function 137 transmits the selected provision model to the central server 200 (step S215). The plurality of sites 100 each transmit a provision model to the central server 200. The central server 200, in the integration function 232, integrates the plurality of provision models transmitted by the plurality of sites 100 to generate a new global model (step S109). The integration function 232 transmits the generated new global model to each site 100 using the communication interface 210 (step S111). In this way, the information processing system 1 ends the processing shown in FIG. 4.

[0060] In the information processing system according to the embodiment, a global model is generated by integrating provision models provided by a plurality of sites 100. As a result, a global model is generated that reflects the characteristics of each of the plurality of sites 100. For example, one site located in a medical facility with a large number of young people provides a provision model based on data that includes a large amount of data obtained from young people, and another site located in a medical facility with a large number of women provides a provision model based on data that includes a large amount of data obtained from women. Therefore, a global model that takes into account the trends in the learning data for each site 100 can be generated.

[0061] Furthermore, in the information processing system of the embodiment, an index value for selecting a model to be provided to the central server 200 is calculated based on the global data distribution (weight) and the local data distribution. Therefore, since an index value that takes the global data distribution (weight) into consideration can be calculated, the accuracy of evaluating the global model generated by the central server 200 can be improved.

[0062] Furthermore, in the information processing system of the embodiment, an index value that takes into account the global data distribution (weight) is used to select a provision model to be provided to the central server 200. Therefore, a model that is close to a global model that is an integration of multiple provision models (global models) provided by multiple sites 100 can be provided to the central server 200.

[0063] In the above embodiment, the information processing system includes the site 100 and the central server 200, but the site 100, which is a collection of the configuration of the central server 200, may function as an information processing device. In this case, the configurations provided at multiple sites may be collected into one site, and training data may be transmitted from the other sites, and a model for the other sites may be generated at one site.

[0064] In the above embodiment, the index value calculation function 136 uses the error included in the global data distribution as the index value, but the index value does not have to be the error and may be, for example, the accuracy. The first selection function 137 selects a model to be provided based on the index value and transmits it to the central server 200, but the first selection function 137 or the index value calculation function 136 may transmit the index value to the central server 200.

[0065] In the above embodiment, the local data distribution calculation function 134 uses the global density ratio as a weight for calculating the index value, but the weight is not limited to the global data distribution. For example, the weight may be the ratio of the data amount between the local data distribution and the global data distribution, or the percentage of missing data.

[0066] In the above embodiment, the index value calculation function 136 calculates an index value that serves as an index for selecting a model to be provided, but for example, it may also calculate an index value of a global model transmitted by the central server 200. In this case, the index value calculation function 136 can evaluate the global model using the calculated index value. The site 100 may transmit the calculated index value of the global model to the central server 200 or another site 100 and provide it.

[0067] In the above embodiment, the global density ratio as a weight is calculated at the site 100, but weights such as the global density ratio may be calculated at a site other than the site 100. For example, the central server 200 has both the local data distribution transmitted by the site 100 and the calculated global data distribution. Therefore, weights such as the global density ratio may be calculated at the central server 200 and transmitted to the site 100.

[0068] 4, the site 100 updates the global model and the local model using the training data stored in the training data storage area 141, and then calculates the global density ratio in step S209. Alternatively, the site 100 may, so to speak, reverse the order of steps S207 and S209 in FIG. 4, calculate the global density ratio before updating the global model, and update at least one of the global model and the local model stored in the memory 140 using at least one of the calculated global model density ratio and each data distribution.

[0069] By updating the global model and local models using the global model density ratio, it is possible to obtain an effect similar to that of so-called covariate shift adaptive learning, in which the density ratio is taken into consideration when calculating the loss function. For this reason, models that have undergone covariate shift adaptive learning can be included as candidates for the model to be provided selected in step S213. Note that, under the assumption that the calculated global density ratio is 100% accurate, a model that has undergone covariate shift adaptive learning would always be selected as the model to be provided, but since errors and the like are actually included, it is preferable to use a model that has undergone covariate shift adaptive learning as a candidate model to be provided.

[0070] According to at least one of the embodiments described above, the accuracy of evaluating a global model can be improved by having a first acquisition unit that acquires a global model based on a dataset related to a plurality of cohorts, a second acquisition unit that acquires a local model based on a first dataset related to at least one first cohort among the plurality of cohorts, a weight calculation unit that calculates a weight based on a first data distribution related to the first dataset and a second data distribution related to a second dataset related to the plurality of cohorts, and an index value calculation unit that calculates an index value related to the global model based on the first dataset and the weight.

[0071] Although several embodiments have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, as well as within the scope of the invention and its equivalents as defined in the claims. [Explanation of symbols]

[0072] 1. Information processing system 100...sites 110,210...Communication interface 120,220...input interface 130, 230...Processing circuit 131...First acquisition function 132...Second acquisition function 133...Learning function 134...Local data distribution calculation function 135...Weight calculation function 136...Index value calculation function 137...First selection function 138...Second selection function 140,240...Memory 141...Learning data storage area 142...Global model storage area 143...Local model storage area 144: Storage area for trained global models 145...Trained local model storage area 150A...First training data 150B...Second training data 160…Local data distribution 180A…First local data distribution 180B…Second local data distribution 200...Central server 231...Global data distribution calculation function 232…Integrated Functions 260,280…Global data distribution 360...Global density ratio NW: communication network

Claims

1. a first acquisition unit that acquires from a central server a global model based on a plurality of datasets relating to a plurality of cohorts and a global data distribution of the plurality of datasets; a second acquisition unit that acquires a first dataset related to a first cohort among the plurality of cohorts and acquires a local model based on the collected first dataset; a weight calculation unit that calculates a weight that is a ratio between a first data distribution of the first data set and the global data distribution; an index value calculation unit that calculates, based on the first dataset and the weights, an index value for each model in a model group including the local model, the global model, and a trained global model obtained by training the global model using the first dataset, to select a provided model to be used in generating a new global model; a first selection unit that selects the provision model to be provided to the central server from the group of models based on the calculated index value; Information processing device.

2. a learning unit that uses the first data set to learn at least one of the global model and the local model; The information processing device according to claim 1 , further comprising:

3. a second selection unit that selects an operating local model to be operated in a medical procedure from a plurality of models including the global model and the local model; 3. The information processing device according to claim 1.

4. a plurality of sites collecting datasets relating to a cohort and obtaining local models based on the collected datasets; a central server that obtains a global model based on the plurality of datasets collected by the plurality of sites; The site: a first acquisition unit that acquires the global model and global data distributions of the plurality of datasets provided by the central server; a second acquisition unit that acquires a first dataset related to a first cohort among the plurality of cohorts and acquires a local model based on the collected first dataset; a weight calculation unit that calculates a weight that is a ratio between a first data distribution of the first data set and the global data distribution; an index value calculation unit that calculates, based on the first dataset and the weights, an index value for each model in a model group including the local model, the global model, and a trained global model obtained by training the global model using the first dataset, for selecting a provided model to be used in generating a new global model; a first selection unit that selects the provision model to be provided to the central server from the group of models based on the calculated index value; Information processing system.

5. The central server An integration unit is further provided to integrate the provision models provided by the plurality of sites into a new global model. The information processing system according to claim 4 .

6. The information processing device obtaining a global model based on multiple datasets for multiple cohorts and a global data distribution for the multiple datasets from a central server; collecting a first dataset for a first cohort of the plurality of cohorts and obtaining a local model based on the collected first dataset; calculating a weight that is a ratio of a first data distribution of the first data set to the global data distribution; calculating, based on the first dataset and the weights, an index value for each model in a group of models including the local model, the global model, and a trained global model obtained by training the global model using the first dataset, for selecting a provided model to be used in generating a new global model; selecting the provided model to be provided to the central server from the group of models based on the calculated index value; Information processing methods.

7. In the information processing device, obtaining from a central server a global model based on a plurality of datasets relating to a plurality of cohorts and a global data distribution for the plurality of datasets; collecting a first dataset for a first cohort of the plurality of cohorts and obtaining a local model based on the collected first dataset; calculating a weight that is a ratio of a first data distribution of the first data set to the global data distribution; calculates, based on the first dataset and the weights, an index value for each model in a group of models including the local model, the global model, and a trained global model obtained by training the global model using the first dataset, for selecting a provided model to be used in generating a new global model; selecting the model to be provided to the central server from the group of models based on the calculated index value; program.

Images